DeepSeek’s Hedge-Fund-Only Era Ends With a Reported $7.4 Billion Round

DeepSeek’s financing story has materially changed: the AI company is no longer presented as relying solely on founder Liang Wenfeng’s quantitative-investment business. Reuters’ July 25 funding update put the completed maiden round at about $7.4 billion and relayed that a second round targeting a valuation of roughly 500 billion yuan had been paused, while noting that Reuters could not independently verify the pause.

By mid-August 2026, the defensible conclusion is narrower than either “fully self-funded” or “conventionally venture-backed.” High-Flyer remains important because it gave DeepSeek capital, technical expertise and room to pursue research before outside investors arrived, but the first external round means it is no longer the whole financing story.

The original High-Flyer advantage was real

DeepSeek emerged from an unusual combination of quantitative finance and artificial-intelligence research. Liang had already built High-Flyer around algorithmic investment strategies, creating access to engineers, computing infrastructure and privately controlled capital without first assembling a conventional venture syndicate.

That arrangement could support long research cycles and reduce pressure to demonstrate immediate commercial returns. It also concentrated strategic control: fewer outside shareholders meant fewer competing demands over budgets, model releases and the timing of monetisation.

The advantage, however, was never equivalent to an unlimited corporate bank account. Assets managed by a hedge fund belong to its investment vehicles and clients, while the manager’s usable income depends on fees, ownership, expenses, redemptions and applicable rules. High-Flyer’s scale therefore indicates financial leverage for Liang, not a dollar-for-dollar pool available to DeepSeek.

A strong fund does not disclose DeepSeek’s budget

High-Flyer entered 2026 from a position of considerable strength. Bloomberg’s account of its 2025 performance placed average returns across the funds at 56.6% and assets under management above 70 billion yuan, using data compiled by Shenzhen PaiPaiWang Investment & Management.

Those figures show that Liang remained connected to a large and profitable financial platform as DeepSeek moved toward outside capital. They do not establish how much money passed from the asset manager, its owners or related entities into the AI operation.

This distinction rules out a common shortcut: applying a standard management-and-performance-fee formula to total assets and labelling the result DeepSeek’s “war chest.” Without the actual fee schedules, ownership economics, expenses and transfers, that calculation produces an estimate of a hypothetical fund manager’s revenue—not evidence of DeepSeek’s available cash.

The more supportable assessment is that High-Flyer gave DeepSeek financing flexibility and reduced its early dependence on external investors. Its successful year may have strengthened that position, but available disclosures do not provide a complete map of the financial relationship between the two businesses.

The $5.576 million figure answers a different question

The financing narrative is often mixed with the claim that DeepSeek built a frontier model for less than $6 million. That number has a documented basis, but it describes a specific computing-cost estimate rather than the expense of establishing and operating the company.

The DeepSeek-V3 technical report calculates $5.576 million for the official training run from 2.788 million Nvidia H800 GPU hours at an assumed rental price of $2 per GPU hour, and explicitly excludes prior research and architectural, algorithmic and data-related experiments.

The estimate therefore excludes costs such as earlier unsuccessful work, employee compensation, data preparation, hardware ownership, inference infrastructure and service operation. It is evidence of engineering efficiency within a defined training configuration, not a disclosure of DeepSeek’s total model-development spending or corporate budget.

This explains why an efficient laboratory can still seek billions in external financing. Training a successful model is only one part of the cost structure; serving users, expanding computing capacity, retaining researchers and running repeated experiments can require far more capital over time.

Outside capital changes the structure, not the technical thesis

The maiden round does not demonstrate that DeepSeek’s efficiency claims failed. It shows that model-level efficiency and company-level financing measure different things: one concerns computing resources used for a defined workload, while the other determines how much infrastructure and research the organisation can support across multiple model generations.

Outside investors increase potential spending capacity but also introduce economic interests beyond Liang and affiliated entities. The consequences depend on rights that are not fully visible in the public record, including voting power, transfer restrictions, information rights and the conditions attached to future liquidity.

This makes ownership and governance more consequential than the headline amount alone. Capital routed through an investment vehicle can provide economic exposure without giving every participant direct control over the operating company, while a strategic or state-backed investor may receive different terms.

The proposed valuation for the second round should likewise be treated as a negotiation target rather than completed financing. A fundraising discussion can be delayed, revised or abandoned before agreements are signed, so it does not add to the company’s available capital merely because a valuation has circulated.

High-Flyer is now one pillar of a mixed model

DeepSeek’s original financial edge was the ability to begin ambitious AI research without waiting for conventional technology investors. High-Flyer supplied an established business base, quantitative expertise and an unusually patient source of affiliated support at a formative stage.

The first outside round changes that description without erasing the advantage. DeepSeek now fits a mixed model: founder influence and hedge-fund resources remain important, while external investors provide additional capacity for a capital-intensive expansion.

The “secret weapon” is no longer hidden funding alone. DeepSeek’s financial position now rests on the interaction between High-Flyer’s resources, documented engineering efficiency and outside capital. Its next test is whether that larger financing base can preserve the research freedom that made the original structure distinctive.

Also read:

• BlackRock Halts Full Redemptions in $26 Billion Private Credit Fund Amid Investor Panic
• Peter Thiel’s Founders Fund Backs AI Cow Collar Startup Halter at $2 Billion Valuation
• Kling AI Raises $2.8 Billion at $18 Billion Valuation in July 2026 Round